background contextual information

 stated or implied methodologies

 explicitly stated assumptions

 structural features

 key concepts and categories.

Background

The upper-level ontologies surveyed are developed within academic or joint academic-government initiatives, sometimes as part of larger projects. For example, the BFO ontology has been developed within the Institute for Formal Ontology and Medical Information Science (IFOMIS), utilising a grant to ‘develop a formal ontology that will be applied and tested in the domain of medical and biomedical information science’ (IFOMIS 2007). Similarly the DOLCE ontology has been developed as part of ‘EU 1ST integrated project Semantic Knowledge Technologies (SEKT)’, a project funded by the ‘EU 6 Framework programme’ (Semantic Knowledge Technologies 2007). Two of the ontologies—BFO and GFO—have been developed with specific focus on medical classification applications. The institutional nature of ontology engineering is common to domain-specific ontologies also, although several popular ontologies have been developed in the public domain, without any notable institutional involvement—the Friend of a Friend (FOAF) ontology is one such example (FOAF 2007).

All of the ontologies have been developed in Europe, with the exception of SUMO, developed in the United States. Each of the ontologies have been presented at semantic web-related conferences—suggesting that, at the time of publication, upper-level ontologies generally have been of greater interest to academic communities than to either commercial or software engineering communities. In addition, there are numerous academic publications which cite these ontologies or the papers describing them, either in the context of ontology engineering specifically, or in relation to broader ontological questions.

Many of the authors of findings actively contribute to public domain mailing lists where ontology engineering issues are discussed. For example, Barry Smith and John Sowa have been actively involved in the construction of two of the featured ontologies (BFO and SUMO respectively), and also have contributed extensively to mailing lists analysed further on in this analysis. In spite of the lack of mailing lists or other discursive sources dedicated to the surveyed ontologies themselves, there is active online debate about the sorts of concepts and distinctions that feature in upper-level ontologies generally. An analysis of these mailing lists is presented in detail below.

At the time of writing it is difficult to measure the relative impact of the ontologies quantitatively. For the purpose of the survey, two sources have been used to indicate the impact:

Results have been collated from Swoogle and Google Scholar searches conducted in October 2007 and October 2009. They are presented in tables 9.1 and 9.2.

The comparison of the searches between Swoogle and Google Scholar show a positive correlation of 0.67 for the results in 2007, and a weaker correlation of 0.37 for those in 2009. With the exception of the large number of ontology references for BFO, DOLCE and SUMO have more ontology references and citations. This does not take into account private usage of these ontologies, but provides a useful heuristic of present adoption rates. It indicates also the increased use of upper-level ontologies over time, according to both of the metrics used, with numbers roughly doubling on average over the two-year period for all ontologies surveyed. As mentioned, the BFO ontology experienced a large surge in the number of ontology references—this abnormal jump appears to be the result of widespread references from newly developed biological ontologies housed by the OBO Foundry, which the BFO was designed to support (Grenon 2003a).

It is also worth noting that there have several prominent efforts at upper-level ontologies that predate the emergence of OWL as a standard for modelling ontologies:

These are expressed in different formalisms, and so are not amenable to algorithmic comparison with those expressed in OWL. The SUMO ontology is explicitly indebted to the SUO ontology (Niles and Pease 2001), and most of the ontologies described have some level of mapping to Word Net, a publicly available online dictionary, which links words based on semantic relations (Miller 1995). Cyc is an older and well-established representation of upper-level and domain-level knowledge (Lenat 1995), developed privately by Cycorp. It is frequently cited in the literature on upper-level ontologies (and knowledge representation), but because of its proprietary nature and different formal structure it is not available for direct comparison.

Finally, as discussed at further length in the section ‘Assumptions’, below, several of the upper-level ontologies have been developed with awareness of competing ontologies, and so reference distinctions between them explicitly. For instance, the GFO ontology is compared with SUMO and DOLCE by its authors (Herre 2009); the BFO ontology is compared with DOLCE (Grenon 2003a); the PROTON ontology references WordNet and Cyc (Terziev, Kiryakov and Manov 2004); and the SUMO ontology, as mentioned, is itself compiled from SUO and a range of other sources:

There has also been a certain amount of post facto literature comparing DOLCE and SUMO in particular, the innovatively entitled ‘DOLCE ergo SUMO’ (Oberle et al. 2007) and ‘OntoMap’ (Kiryakov, Simov and Dimitrov 2001) being two such examples. This activity suggests that ontology engineering is a highly dynamic social process, and at the present time, still far from finding agreement for the abstract concepts upper-level ontologies describe. The engagement with the mailing lists— the informal ‘chatter’ that sits behind the formal austerity of ontologies—explores this dynamic and dialogical process further.

Methodologies

The methodologies outlined in the literature surrounding upper-level ontologies vary from highly explicit (SUMO) to implicit (PROTON). Table 9.3 provides an overview of the methodological approach adopted—as best as can be inferred if not otherwise stated—and the degree of formality of process of ontology construction.

The variance in methodology (in terms of the adopted approach itself, and the degree of explicitness about the approach) is one indicator about the degree of commensurability between upper-level ontologies. How concepts are selected and arranged can divulge further assumptions not explicit in the conceptualisation. However, even in the case of the SUMO ontology, little is stated about how one concept was chosen over another, what leads to a particular arrangement of concepts, beyond an acknowledgement of certain existential assumptions, and what degree of detail is suitable. Arguably the criterion of methodology is harder to apply to upper-level ontologies especially, given the abstractions of the concepts concerned. Subsequent discussion returns to this variance, since it has a bearing on what can be said about the commensurability of these ontologies.

Assumptions

None of the upper-level ontologies surveyed lay great pretensions towards definitiveness; the authors of DOLCE for instance state: ‘we do not intend DOLCE as a candidate for a “universal” standard ontology’ (Masolo et al. 2002). In the literature showcasing the ontologies, their designers are overtly aware of the assumptions which characterise their construction (Masolo et al. 2002). Nevertheless there are clear differences in these assumptions, in kind and degree. The following section examines what assumptions are employed, and how they compare.

Structural features

The most evident structural difference between the ontologies is in size. Table 9.5 compares the number of classes, properties and individuals contained in each of the ontologies.

As indicated, the PROTON and DOLCE ontologies have been constructed as a series of smaller ontologies, which are linked together via the imports construct. In contrast, the other ontologies are contained within single files.

The PROTON ontology has been separated into three separate ontologies: system, top and upper. The upper imports the top ontology, which in turn imports the system ontology. The DOLCE ontology contains eight separate subsidiary and interconnected ontologies, yielding a more complex structure:

To use the SocialUnits ontology, for example, means to import the classes and properties from InformationObjects, ExtendedDnS, TemporalRelations, SpatialRelations and DOLCE-Lite ontologies.

It is also noticeable that the ontologies differ in the degree that they use classes over properties, expressed in the ratio figures above. In the case of DOLCE in particular, this suggests what might be termed a ‘functional’ or ‘attributive’ approach to the organisation of entities, since the majority of its conceptual constructs are properties rather than classes. This point is elaborated further in the discussion of categories below.

BFO

Entity

DOLCE

particular

GFO

Entity

PROTON

Entity

SUMO

Entity

Even this schematic outline shows some initial points of similarity and difference between the five ontologies. Each starts with a common root concept, Entity, or its near-cognate: the DOLCE ontology has a root concept of particular, which it defines as ‘entities which have no instances’ (Masolo et al. 2002), distinguished from universals, which do not adhere to this constraint. The following conceptual distinctions, or near synonyms of them, are also common to at least three of the ontologies:

These distinctions can be mapped to specific concepts in each of the five ontologies; these are depicted in screen shots in the appendix to this chapter.

This comparison demonstrates that there is a high degree of overlap in the use of concepts across the ontology set and, equally, that these concepts are differently configured, so that establishing direct concordances between apparently synonymous concepts is risky. The interpretation of the above distinctions, when applied to particular ontologies, carry considerable ambiguity. For example, the Conceptual–Physical distinction aligns Concept (GFO), [GeneralTerm–Topic] (PROTON) and Proposition (SUMO, DOLCE) as all conceptual entities—clearly, however, they are not all synonymous terms. More tenuously, the meaning of Dependent might appear to be preserved by the [Relation— Attribute—Quantity] (SUMO) and quality (DOLCE) classes, since these have equivalently named terms in other ontologies which are subclasses of kinds of Dependent entities: the DependentContinuant and Dependent classes of BFO and GFO ontologies respectively. The Continuant–Occurrent distinction is also difficult to interpret consistently across the ontologies. Broadly, continuant entities are those which ‘are wholly present (all their parts are present) at any time at which they exist’ while occurrent entities are those ‘that extend in time and are only partially present for any time at which they exist because some of their temporal parts may be not present’ (Bazzanella, Stoermer and Bouquet 2008). The GFO ontology synonym for this distinction, the Continuous-Discrete pair, nevertheless contains a mixture of both object-like and process-like classes within each of the Continuous and Discrete super classes—and the ontology also contains a sibling Presential class, which appears a more natural synonym for Continuant. The BFO ontology subsumes TemporalRegion under Occurrent and SpatialRegion under Continuant classes, but also adds SpatiotemporalRegion as a subclass of Occurrent, with the following extended note:

This confirms the point made earlier: two ontological perspectives are supported by the BFO—one sharply distinguishing spatial and temporal, continuant and occurrent entities, the other collapsing within a general occurrent, spatio-temporal conceptual apparatus. Habituation of two fundamental perspectives within the one ontological housing creates a point of dissonance with other ontologies which maintain a single viewpoint, preserving ‘continuant’ and ‘occurrent’ entities as primordially distinct and mutually disjoint. The SUMO and DOLCE ontologies are largely consistent in maintaining this distinction, while the PROTON ontology seems idiosyncratically to treat JobPosition—arguably a continuous entity—as a kind of Happening, while Service (in the sense of a service rendered) is a kind of Object.

The difficulty of drawing synonymous relations across ontologies is more notable still in a comparison of the summative effect of multiple conceptual distinctions. For example, while each of the ontologies has classes to represent the Spatial-Temporal and Continuant-Occurrent conceptual pairs, how these distinctions are organised differs markedly. The BFO ontology makes a primary distinction between Continuant and Occurrent classes. The distinction between spatial and temporal entities is then a subordinate one—more formally, the pair [SpatialRegion/TemporalRegion] is subordinate to the pair [Continuant/Occurrent]. The GFO ontology treats these distinctions as equivalent—both [Discrete/Continuous] and [Space_time/Space and Space_time/Time] class pairs are subsumed by the Entity/Item/Individual class. The PROTON ontology follows the BFO organisation: the pair [Object/Location and Happening/Timelnterval] is subordinate to the [Object/Happening] pair. Meanwhile, the SUMO ontology has an orthogonal relation between the equivalent pairs: while Object/Region is clearly subordinate to Object, the nearest temporal synonym, TimeMeasure, is treated as a kind of Quantity, not a subordinate of Process—the nearest SUMO synonym to the Occurrent concept used by the BFO. The DOLCE ontology pairs both sets of conceptual classes: the Continuant-Occurrent distinction is matched by the near-synonymous pair [endurant/perdurant], while the Spatial-Temporal distinction is matched by the [space-region/temporal-region] pair of classes. However, the latter pair is subordinate to a higher level class, abstract, which is distinguished from the spatio-temporal-particular class. Instances of the spatio-temporal-particular class (whether members of an endurant, perdurant, or of another class) are bound to space or time regions via properties, rather than through direct class subsumption relations. This suggests conversion between DOLCE and other ontologies would need to interpret and transform other kinds of relations into class subsumption ones.

Similar complexities can be found in other conceptual overlays. The BFO ontology has no class pair corresponding to the Abstract-Concrete distinction, yet synonymous class pairs are primary distinctions for the PROTON, SUMO and DOLCE ontologies (and a more subordinate distinction for the GFO ontology). Only one of the ontologies, DOLCE, reinforces this conception distinction with a logical constraint: the abstract class is declared logically disjoint from endurant, perdurant and quality subclasses of the spatio-temporal-particular class, and further annotates the abstract class: ‘The main characteristic of abstract entities is that they do not have spatial nor temporal qualities, and they are not qualities themselves’ (Gangemi 2006).

The five ontologies also have clear conceptual areas of what might be termed ‘perspectival specialisation’—areas in which they move beyond upper-level abstractions towards domain-level specificity. The BFO ontology, with 36 classes and no properties, contains only abstract physical classes, roughly representative of the distinctions introduced above (Continuant–Occurrent, Dependent–Independent, Collective-Individual). The emphasis on supporting the mutually exclusive three-and four-dimensional perspectives is the main unusual feature of its categorial structure—although this feature is also supported by DOLCE and SUMO ontologies within a single perspectival view. The GFO ontology uses a number of specialised terms further down the class hierarchy which indicate a scientific or technical orientation: Chronoid (‘Every chronoid has exactly two extremal and infinitely many inner time boundaries which are equivalently called time-points’), Topoid (‘connected compacted regions of space’), Configuroid (‘integrated wholes made up of material structure processes and property processes’) and Situoid (‘processes whose boundaries are situations and which satisfy certain principles of coherence, comprehensibility, and continuity’) are examples (Herre 2009). The PROTON ontology contains Product, Service, Document, JobPosition and Contactlnformation, which relate more specifically to organisational or commercial fields (Terziev, Kiryakov and Manov 2004). Moreover PROTON makes a first-order distinction between Entity, LexicalResource and EntitySource, suggesting a primary demarcation between entities in the world, and their discursive description and provenance.

The SUMO ontology explicitly incorporates the intentional or constitutive standpoint of an agent—a large number of agent actions, including Guiding, Classifying, Listening, Looking and Meeting, are subsumed within an intentionalProcess class. These total 114 classes, or 20 per cent of the number of classes in the ontology as a whole, which is suggestive of an internalist or subjectively-oriented perspective towards ontological entities. Finally, the DOLCE ontology largely mirrors the kinds of distinctions maintained by SUMO, but models these as object properties between classes, rather than as class inheritance. Thus it sees entities as bound by mereological and functional, rather than by subsumption relations. This formalist perspective is mirrored by object property names such as parameterises, postconditions, preconditions, deputes and interprets. The SUMO ontology has a similarly large number of object properties, but uses more colloquial and lay terms such as causes, employs, larger and uses.

Table 9.6 summarises these underlying differences in orientation. These qualitative distinctions could be used to further form the basis of a set of quantitative valuations against the dimensions introduced in the commensurability framework chapter, as well as refinement of what those dimensions, in the case of the five upper-level ontologies, should be; at this stage, it is enough to demonstrate the utility of the kind of interpretative analysis employed here for teasing out what some of the salient distinctions between the ontologies are. The next section examines the sociological context in which these distinctions are voiced and debated, through an online content analysis of two mailing lists. This analysis also brings forward further suggestive extrinsic or social distinctions which mark these ontologies.

I think we’re arguing about the definitions of our terms, here. My use of the term ‘Truth’ causes cognitive dissonance for you.

Well, you haven’t actually defined it: but I think I get your drift. It doesn’t cause me cognitive dissonance (if it did, I might be more inclined to agree with it): I just think its mistaken (Ontolog Forum 2010, message 188).

Analysis of mailing lists

To analyse the lists, a small software script was developed to harvest the contents of emails from their publicly available archives (Ontolog Forum 2010; W3C 2010). The script retrieved posts since their inception (March 2000 for the Semantic Web Interest Group and May 2002 for Ontolog Forum) until 18 May 2009. Mailing list archives typically employ the following structure:

The script exploits this common structure (at least for the two archives in question) and, starting with the index page, follows links to the date and message pages automatically. On each of these pages it parses the contents for common elements, such as the date, author, subject and contents of the message. Importantly, message ‘threads’ retain the same subject heading for the most part, permitting analysis of common topics and keywords. The script then captures unique authors, subjects, messages and individual words in a database system. Care has been taken to ensure accuracy in the results; however, for various reasons, it is difficult to gain precision with subject, author and word counts—continuous subjects can be arbitrarily varied by authors or their mail clients; authors can post from different mail accounts, with different names or variants; word morphographic variants, aside from plurals, can be difficult to correlate automatically; email thread subject headings can be renamed by correspondents, to follow new ‘threads’ of conversation; and mailing lists can allow for the possibility of ‘spoofing’, faking email contents, due to lack of rigorous authentication methods. The last concern, given the specialised nature of these lists, is a relatively low risk in these particular cases.

Quantitative analysis

Table 9.7 summarises the number of messages received on the Semantic Web Interest Group and Ontolog Forum between 2000 and 2009. These figures show that there has been a rising interest in the forums in recent years—the Semantic Web list grew rapidly in 2005, the Ontolog Forum grew in 2007. The total number of messages is comparable, with indications in 2009 that Ontolog Forum experienced roughly twice the amount of activity.

A count of authors and subjects suggests that Ontolog Forum has a smaller and more focused community than the Semantic Web Interest Group (Table 9.8). These numbers indicate that each author contributes on average approximately 23 messages across the entire period surveyed (2000 through to May 2009), and each subject receives four messages on

Ontolog Forum, compared with approximately eight and two messages respectively received on the Semantic Web list. This is confirmed by an analysis of the top 20 authors and subjects, which show higher message-to-author and message-to-subject ratios for Ontolog Forum.

Although the five ontologies and two comparative studies (OntoMap and SWIntO) analysed above are products of private development, and consequently do not have mailing lists or other public fora, several of the contributing authors featured in the lists surveyed. Table 9.9 shows each of the contributors featuring in the lists, along with the number of messages posted.

These figures can be used as basic heuristics for some of the dimensions introduced for classifying the ontologies above. While overextrapolation from these figures can be misleading—for instance, other relevant mailing lists are not included here, and there is no means of reviewing offline discussions—based on the characterisation of the lists themselves, it is possible to make several inferences:

These tentative observations are corroborated by a word frequency analysis of the ontology terms themselves. Fortuitously, each of the ontologies has an acronym that is unusual enough to make collision with quotidian usage quite unlikely. Table 9.10 shows the number of times each ontology is mentioned on the Semantic Web Interest Group and Ontolog Forum, along with their frequency relative to the most commonly cited ontology. Discussion on Ontolog Forum is predictably far more prolific, as it is dedicated to the establishment of foundational or upper-level ontologies. All of the ontologies are mentioned more often in absolute terms on Ontolog Forum, with SUMO, DOLCE and BFO cited more often by a factor of ten or more. Notwithstanding this absolute difference, in relative terms DOLCE is mentioned twice as often on the Semantic Web Interest Group list as SUMO, while SUMO is mentioned nearly three times more often on Ontolog Forum. Both GFO and PROTON are mentioned relatively infrequently, reinforcing earlier suggestions that these ontologies have low levels of interest and community engagement.

Qualitative analysis

One conversational thread from Ontolog Forum has been selected for more detailed qualitative analysis. Entitled ‘Two ontologies that are inconsistent but both needed’, the conversation involves one of the authors of the BFO ontology (Barry Smith), and other influential participants—the author of the RDF specification, Pat Hayes; a frequently cited author on ontologies, John Sowa; and a number of contributors to various standards initiatives. Usefully, the subject matter covers both the general problem of commensurability between ontologies, as well as the specific question of interoperability between upper-level ontologies. The conversation takes place in June 2007, some time after the ontologies covered here were developed, so it does not directly relate to the background of their development. Nevertheless it elucidates many of the foundational issues involved in upper-level ontology engineering, which remain active subjects for debate, and brings into view a range of perspectives on the challenges of interoperability between multiple ontologies.

Commensurability of upper-level ontologies

The exploration of the five ‘upper-level’ ontologies has suggested considerable areas of both similarity and difference. Structurally, the PROTON and DOLCE ontologies show greater modularisation, while PROTON, DOLCE and SUMO are considerable larger and semantically denser than GFO and BFO. The DOLCE ontology also favours use of object properties to relate entities functionally, rather than via class subsumption relations. This is carried over in the more scientific and theoretical orientation of DOLCE, evidenced by use of specialised terminology, a feature it shares with GFO, and to some extent with BFO. Comparatively, the PROTON and SUMO ontologies share a more pragmatic and vernacular orientation. The BFO, SUMO and DOLCE also permit multiple perspectives on physical entities. They can be described using spacetime coordinates (4D), or with clearly demarcated spatial and temporal characteristics (3D), along a more abstract distinction between ‘Continuant’ and ‘Occurrent’ categories. This is evident in the BFO ontology directly, but requires some reference to the surrounding literature for SUMO and DOLCE. It is unclear how the GFO and PROTON ontologies are positioned around this distinction, but since both support some variant of the Continuant-Occurrent distinction, it can be assumed they operate within a three-dimensional paradigm. The SUMO ontology, and to a much lesser degree the PROTON and DOLCE ontologies, also explicitly model an agent’s intentional relation to entities in the world, permitting—though not necessarily insisting on—a constructivist rather than naturalist outlook. Despite considerable overlap, conceptual equivalence, or synonymy, is frequently hard to establish, because of the differing levels of terminological intersection, use of functional roles over class subsumption (in the case of DOLCE), greater conceptual density with the larger ontologies (in the cases of PROTON, SUMO and DOLCE) and a general lack of transparent isomorphisms between the conceptual graphs of the ontologies.

Table 9.12 summarises some of the findings of the exploration of the five ‘upper-level’ ontologies, using a combination of the dimensions introduced in the general model in Chapter 12, ‘A framework for commensurability’, and those which have presented themselves in the course of the analysis of upper-level ontologies particularly. In addition, I have added the SWIntO variables (Oberle et al. 2007) as supplementary dimensions, since these are largely specific to the upper-level ontologies described, and are not included in my general taxonomy. Valuations for each of the ontologies are relative—a low valuation on the ‘small vs. large’ dimension, for example, indicates a small number of logical axioms relative to the other ontologies considered. ‘Low’ and ‘high’ values reflect evaluations against the second term of each conceptual opposition expressed by a dimension.

From this matrix several patterns emerge. SUMO and DOLCE ontologies match up on a number of dimensions, and do not differ greatly on any. That they are broadly commensurable is further borne out in efforts to develop translations between them (Oberle et al. 2007). However, their relative size and complexity would suggest a large number of ‘local’ commensurability issues, at particular branches of their respective taxonomic structures. BFO seems to differ markedly from both of these, and indeed from the aims of PROTON as well. Since it is similarly designed for use in scientific and biological systems, it is perhaps unsurprising that the GFO ontology is closest, at least for dimensions where values can be meaningfully derived. Otherwise, the GFO and PROTON ontologies stand out as relatively idiosyncratic in terms of usage and available documentation—relatively little more can be inferred from the sources available. A large number of ‘Unknowns’ for these ontologies might imply either better commensurability—or, more likely, since they leave open large degrees of interpretative scope for different domain ontologies importing them, might imply ‘hidden’ pockets of potential incommensurability which might emerge only on further analysis. By contrast, the explicitness attached to BFO, SUMO and DOLCE ontologies suggests that areas of incommensurability are easier to locate up front.

While the mailing list analysis can be used to generate a series of evaluations of the ontologies themselves, this kind of exercise is perhaps more helpful for considering the question of commensurability of upper-level ontologies generally. In particular, the distinctions raised at the end of the analysis are perhaps questions more to be asked of the situational context in which upper-level ontologies are being considered. In some situations it might be useful to ask whether there are metaphysical issues important for the users, systems and requirements at hand; whether different systems must agree on their definition of abstract concepts; whether multiple viewpoints can be accommodated; and whether differences in viewpoints can be negotiated and, if so, how? In other situations, where an analyst needs to evaluate upper-level ontologies, the dimensions and distinctions outlined above could become a series of evaluative criteria. Here it might be useful to ask what it would mean to view the world as ontologically divided into objects and processes, or, on the other hand, to see all things as possessing ‘object-like’ and ‘processlike’ features in different measures—or, pace Smith, to ask whether both viewpoints can be housed in a single, logically consistent but philosophically incommensurable system. A further question might ask whether, from a procedural point of view, these kinds of distinctions ought to arise organically, through piece-meal observation and analysis, or alternatively ought to be imposed as a set of guiding metaphysical assumptions from above. And, finally, it might be asked what kinds of downstream commensurability issues arise by the assumptions made by upper-level ontologies, and the lower-level domain ontologies they are designed to support.

In short, these questions could be more useful as broader interpretive ‘framings’ for consideration of other, more fine-grained commensurability dimensions, such as whether the ontologies employ similar design methods. Consequently, while these dimensions are included under the ‘Perspective’ group in Table 9.12, they better reflect general aspects of the situation in which the ontologies are considered, rather than the ontologies themselves. This point is revisited briefly in the next section.

Implications for a general theory of commensurability

What do these findings suggest for a theory of commensurability? As indicated in Chapter 12, ‘A framework for commensurability’, commensurability is assessed against a contextual backdrop of a given scenario—typically, a project with requirements, aims and purposes. Since the development of upper-level ontologies is what might be described as a niche market, it is hard to characterise the sociological environment in which these ontologies are developed and used. However, the quantitative and qualitative analysis of two mailing lists—featuring some of the ontology contributors themselves, and many other developers and users of ontologies—are revealing. On the one hand, comparatively little discussion takes place over the actual categories of upper-level ontologies, in spite of the avowed purpose of the lists. On the other, the lists do discuss questions of mechanics—how to develop and deploy ontologies— and those of logical entailment within and between ontologies. The Semantic Web Interest Group list is on the whole dedicated to the technological and operational side of ontologies—announcements about conferences, workshops, specification releases, and so on—while Ontolog Forum focuses more on issues of ontology content, but nonetheless features many discussions around broadly logical and methodological issues, rather than those of substantive ontological content. As the thread described above shows, this coverage includes the very problem of ontology commensurability itself.

The lists proved useful as heuristic aids for understanding the general backdrop against which the surveyed ontologies can be understood. One immediate result is that what appear to be posited categories within those ontologies are still very much the subject of considerable contention. Further, what appear to be circumstantial deviations between ontologies—fortuitous reliance on one distinction over another—can be shown to be rooted in fundamental philosophical positions, which are not clearly evident in either the ontologies themselves, or their supporting literature. How vital differences in these positions are again is a matter for context—but clearly, in the case of BFO ontology, one of the key authors holds a resolute view as to the primacy of the top-level categories posited. Drafting connections from BFO to other ontologies based on purely lexical considerations is likely to obscure at least the authorial intent if not the actual extension of these categories to other domain-level ontologies in practice. How much adoption of an upper-level ontology implies a wholesale commitment to its claims is yet a further question, bearing here on the commensurability relation between importing domain-level and imported upper-level ontology. Concern over this question has, in turn, led some members of these lists to advocate either forms of ontological pluralism—in which multiple, incompatible conceptual schemes are happily co-opted—or abnegation—where talk of ‘upper levels’ is ignored altogether. Each of these options implies some weaker form of interoperability, where total agreement is passed over in favour of partial and local—but perhaps workable—agreements.

Practically, the absence of direct background material about the ontologies themselves led me to examine two mailing lists where the peculiar fusion of speculative philosophical and detailed technical subjects are discussed. This suggests that a range of approaches and data sources need to be considered when looking for commensurability ‘clues’. Another consequence of this analysis was the addition of several further distinguishing dimensions to the commensurability evaluation model. However, these dimensions could relate better to the situational context in which commensurability is evaluated, rather than to the ontologies themselves—and suggests that, in terms of the commensurability model outlined in Chapter 12, ‘A framework for commensurability’, it may be important to model the context more stringently. In practice, the vagueness and open-endedness of ‘context’ make such a formal treatment a difficult prospect—despite some existing efforts to do this. I return to this subject in the conclusion to the study, as a candidate area for further refinements and research of the model.

Although the upper-level ontologies surveyed demonstrate impressive theoretical coherence, engaging with the public fora in which they are produced shows that for many questions of fundamental ontological importance—in philosophical and engineering senses—no kind of Platonic-level resolution is in sight. Rather the mailing lists evoke an atmosphere of endless dialogue, intractable positions and an endless recasting of distinctions. Nevertheless, engagement with these social networks provides a necessary warning against any superficial reconciliation between concepts which a purely technical review of ontologies might suggest. It indicates that the very question of commensurability itself is enmeshed within the broader, messy, retail world of discursive practices. By engaging with these practices, it is possible to succumb to a potentially interminable dialogue, far removed from the pristine conceptualisations of ontologies themselves, in which semantic equivalences and distinctions are endlessly debated. Conversely, that engagement reveals the contrived and fictional concordance, the illusory character of the harmonious, but purely formal translation of concepts. It serves as a reminder that knowledge systems are merely the latest emergent suburban region within the much broader metropolis of epistemic discourse.

Upper-level ontologies in Protégé

Conceptual distinctions between the ontologies

Word frequency analysis details

Aristotle. The Internet Classics Archives. Categories, E.M. Edghill (tr.) http://classics.mit.edu/Aristotle/categories.html, 1994. [(accessed 20 March 2010)].

Bazzanella, B., Stoermer, H., Bouquet, P., Top Level Categories and Attributes for Entity Representation Technical Report 1. University of Trento, Scienze della Cognizione e della Formazione. 2008.

BFO. Basic Formal Ontology (BFO). website http://www.ifomis.org/bfo, 2007. [(accessed 22 July 2007)].

Brandom, R. Making It Explicit. Cambridge, MA: Harvard University Press; 1994.

Ding, L., Finin, T., Joshi, A., Pan, R., Cost, R.S., Peng, Y., Reddivari, P., Doshi, V., Sachs, J., Swoogle: A Search and Metadata Engine for the Semantic Web. CIKM 2004: Proceedings of the Thirteenth ACM International Conference on Information and Knowledge Management. New York: Association for Computing Machinery. 2004:652–659.

FOAF. The Friend of a Friend (FOAF) Project. website http://www.foaf-project.org, 2007. [(accessed 22 July 2007)].

Gangemi, A. The DOLCE and DnS ontologies. http://www.loa-cnr.it/DOLCE.html, 2006. [(accessed 22 February 2010)].

Gennari, J.H., Musen, M.A., Fergerson, R.W., Grosso, W.E., Crubézy, M., Eriksson, H., Noy, N.F., Tu, S.W. The Evolution of Protégé: An Environment for Knowledge-Based Systems Development. International Journal of Human-Computer Studies. 2003; 58:89–123.

Grenon, P., BFO in a Nutshell: A Bi-Categorial Axiomatization of BFO and Comparison with DOLCE technical reportInstitute for Formal Ontology and Medical Information Science. Leipzig: University of Leipzig, 2003.

Grenon, P., Spatio-Temporality in Basic Formal Ontology (BFO) technical reportInstitute for Formal Ontology and Medical Information Science. Leipzig: University of Leipzig, 2003.

Grenon, P., Smith, B. SNAP and SPAN: Towards Dynamic Spatial Ontology. Event-Oriented Approaches in Geographic Information Science. 2004; 69:137–171.

Hayes, P., Hayes, P. RDF Semantics. W3C recommendation, World Wide Web Consortium http://www.w3.org/TR/rdf-mt/, 2004. [(accessed 20 January 2010)].

Hayes, P., Patel-Schneider, P.F., Horrocks, I. OWL Web Ontology Language Semantics and Abstract Syntax. W3C recommendation, W3C http://www.w3.org/TR/owl-semantics/, 2004. [(accessed 20 January 2010)].

Herre, H. General Formal Ontology (GFO): A Foundational Ontology for Conceptual Modelling. In: Poli, R., Obrst, L., eds. Theory and Applications of Ontology, vol. 2. Berlin and Heidelberg: Springer; 2009.

Horridge, M. Protégé OWLViz. accessed10February2010, 2005.

Horridge, M., Bechhofer, S., Noppens, O., Igniting the OWL 1.1 Touch Paper: The OWL API3rd OWL Experiences and Directions Workshop (OWLED 2007). Berlin: Springer, 2007.

IFOMIS. IFOMIS. website http://www.ifomis.org, 2007. [(accessed 22 July 2007)].

Kiryakov, A., Simov, K.I., Dimitrov, M., OntoMap: Portal for Upper-Level Ontologies. FOIS 2001: Proceedings of the International Conference on Formal Ontology in Information Systems. New York: Association for Computing Machinery. 2001:47–58.

Lenat, D.B. Cyc: A Large-Scale Investment in Knowledge Infrastructure. Communications of the ACM. 1995; 38:33–38.

Lenat, D.B., From 2001 to 2001: Common Sense and the Mind of HALHAL’s Legacy. Cambridge, MA: MIT Press, 2001. [193–209].

Masolo, C., Borgo, S., Gangemi, A., Guarino, N., Oltramari, A., Schneider, L., The WonderWeb Library of Foundational Ontologies technical report. Institute of Cognitive Sciences and Technology. 2002.

Miller, G.A. WordNet: A Lexical Database for English. Communications of the ACM. 1995; 38:39–41.

Niles, I., Pease, A., Towards a Standard Upper Ontology. In FOIS 2001: Proceedings of the International Conference on Formal Ontology in Information Systems. New York: Association for Computing Machinery. 2001:2–9.

Oberle, D., Ankolekar, A., Hitzler, P., Cimiano, P., Sintek, M., Kiesel, M., Mougouie, B., Vembu, S., Baumann, S., Romanelli, M. DOLCE ergo SUMO: On Foundational and Domain Models in SWIntO (Smartweb Integrated Ontology). Journal of Web Semantics: Science, Services and Agents on the World Wide Web. 2007; 5:156–174.

Ontolog Forum. Ontolog Forum Mail Archives. http://ontolog.cim3.net/forum/, 2010. [(accessed 18 January 2010)].

Searle, J.R. Intentionality: An Essay in the Philosophy of Mind. Cambridge, UK: Cambridge University Press; 1983.

Semantic Knowledge Technologies. Welcome to SEKT—SEKT Portal. http://www.sekt-project.com, 2007. [(accessed 22 February 2010)].

Smith, B. Beyond Concepts: Ontology as Reality Representation. In: Varzi A., Vieu L., eds. FOIS 2004: Proceedings of the Third Conference on Formal Ontology in Information Systems. Amsterdam: IOS Press; 2004:73–84.

Smith, B. Against Fantology. In: Reicher M.E., Marek J.C., eds. Experience and Analysis. Vienna: HPT & ÖBV; 2005:153–170.

Smith, B., Grenon, P. The Cornucopia of Formal-Ontological Relations. Dialectica. 2004; 58:279–296.

Terziev, I., Kiryakov, A., Manov, D., Base Upper-Level Ontology (BULO) Guidance technical report. Semantic Knowledge Technologies (SEKT). 2004.

W3C. Semantic Web Interest Group Mail Archives. World Wide Web Consortium http://lists.w3.org/Archives/Public/semantic-web/, 2010. [(accessed 19 February 2010)].